集成多特征信息的运动阴影检测

Moving shadow detection by integrating multiple features

摘要:

为实现监控场景中运动目标和阴影的准确分割,提出了一种基于GMM和MRF的运动阴影检测与消除算法.首先,利用GMM的学习能力建立背景统计模型并得到前景区域像素集合.其次,将前景区域与对应背景区域间的颜色、边界、纹理和时空一致性等特征信息集成到马尔可夫随机场能量函数中,并利用图割算法实现马尔科夫随机场能量函数的最小化,得到最终的分割结果.最后,在室内和室外不同场景类型视频序列上验证了算法的有效性.实验结果表明,算法在运动阴影检测与消除方面较以往方法具有较好的准确性、可靠性和鲁棒性.

Abstract:

To segment moving objects from their shadows in video surveillance systems,a GMM and MRF-based moving cast shadows detection and removing method was proposed.First,the Gaussian mixture model was adopted to build statistical models to describe background,and the foreground pixels were obtained by background subtraction method.Second,the feature information of color,edge,texture and spatiotemporal coherence between the foreground pixel area and the corresponding background area were integrated into Markov random fields’ energy function.Graph Cut algorithm was used to minimize the energy function,and the final segmentation result was got.Finally,the effectiveness of the method on different video sequences of indoor and outdoor scenes was verified.Experiment results demonstrated that,compared with previous methods,the algorithm could detect and remove moving cast shadows more accurately,reliably and robustly.